Nighttime Cold Stress and Cardiovascular Emergency Hospitalizations: A Time‐Series Analysis by Sex and Age in a Subtropical City

Abstract

Background

Cold temperature can increase cardiovascular risks even in subtropical regions, where infrequent exposure and limited infrastructure may limit cold adaptation. Risks may intensify at night because of declining temperature and reduced thermoregulatory capacity during sleep, yet the impact of nighttime cold exposure remains underexplored.

Methods

A time‐series ecological study using quasi‐Poisson regression and distributed lag nonlinear models assessed associations between cold exposure and 417 044 emergency hospitalizations for cardiovascular diseases, including acute myocardial infarction, congestive heart failure, and cerebrovascular diseases), in Hong Kong (2012–2022). Cold days and nights were defined as periods with hourly temperature below the 5th percentile of daytime maximum or nighttime minimum. Cold intensity was quantified using cold day excess and cold night excess; models were adjusted for air pollution, relative humidity, and temporal trends.

Results

Cold night excess was associated with stronger hospitalization risks than cold day excess across all cardiovascular outcomes, with the highest risks for congestive heart failure, followed by acute myocardial infarction and cerebrovascular diseases. Relative risks for cold night excess versus cold day excess were 1.44 (95% CI: 1.40–1.48) versus 1.19 (1.18–1.21) for congestive heart failure, 1.26 (1.22–1.30) versus 1.12 (1.10–1.30) for acute myocardial infarction, and 1.06 (1.04–1.08) versus 1.03 (1.02–1.03) for cerebrovascular disease (all P<0.01). From 2012 to 2022, 201 cold nights were linked to 6805 (6045–7536) excess cardiovascular disease hospitalizations. Relative risks increased with age, especially among those aged ≥90 years.

Conclusions

Public health strategies need to address not only heat but also nighttime cold stress. Targeted interventions are crucial to mitigate risks in subtropical cities.

Nonstandard Abbreviations and Acronyms

CDE

cold day excess

CNE

cold night excess

DLNM

distributed lag nonlinear models

PM_2.5

fine particulate matter

Clinical Perspective.

What Is New?

• To our knowledge, this study is the first to comprehensively assess the short‐term associations between nighttime cold exposure and cardiovascular emergency hospitalizations in a subtropical city. The relative risk for nighttime cold exposure was 1.44 (1.40–1.48) for congestive heart failure compared with 1.19 (1.18–1.21) for daytime exposure. Older adults were particularly vulnerable across all cardiovascular outcomes during nighttime cold exposure.

What Are the Clinical Implications?

• Clinicians in subtropical regions should advocate for targeted interventions, such as enhanced home insulation and early warning systems for cold nights, to protect vulnerable groups, particularly older adults, while hospital administrators can use predictive modeling to prepare for surges in cardiovascular admissions during cold nighttime weather.

Cardiovascular disease (CVD) is the leading cause of death and disability worldwide. ¹ , ² , ³ , ⁴ According to the GBD (Global Burden of Disease) study, CVD caused approximately 19 million deaths globally in 2021. ⁵ This number is projected to rise to 23.3 million by 2030. ⁶ , ⁷ Beyond well‐established environmental determinants such as air pollution and heat, cold exposure has also been identified as a key environmental risk factor contributing to acute cardiovascular events, with several studies linking low temperature to increased risks of conditions such as acute myocardial infarction (AMI) and heart failure. ⁸ , ⁹ As climate extremes become more frequent and severe, quantifying the association between cold exposure and specific acute cardiovascular conditions has become a critical priority for advancing environmental epidemiology and guiding public health interventions.

Cold exposure has been widely documented to increase the risk of CVD through physiological mechanisms such as vasoconstriction, elevated blood pressure, and increased cardiac workload, which can contribute to myocardial ischemia and acute coronary artery plaque destabilization. ¹⁰ A growing number of epidemiological studies have investigated the health burden of cold exposure across various global regions. For example, a large‐scale multi‐country analysis found that extreme cold was significantly associated with increased mortality from several cardiovascular diseases. ¹¹ In China, cold conditions such as low ambient temperature and cold spells have been linked to higher risks of AMI. ¹² However, most of these studies have focused on temperate or cold climate regions, with limited attention given to subtropical cities, where residents are more likely to exhibit a narrower range of thermal adaptability to cold compared with those in cold climates. ¹³ Evidence from Hong Kong, for example, shows no significant heat‐related increase in AMI risk but reveals a strong association between cold exposure and AMI hospitalizations, suggesting that people living in subtropical regions may be more resilient to heat but particularly susceptible to cold. ¹⁴ Building on this evidence, this study broadens the disease scope beyond AMI to include congestive heart failure (CHF) and cerebrovascular diseases, as these conditions are also exacerbated by cold‐induced vasoconstriction and elevated blood pressure, which can precipitate heart failure exacerbations and stroke. ¹⁵ This approach enables a more comprehensive assessment of cold‐related cardiovascular risk.

While the health risks of cold exposure are well‐documented, research on the specific associations with nighttime cold exposure is limited. Cold conditions may impose greater physiological stress during nighttime compared with during daytime, as the circadian drop in core body temperature during sleep, combined with lower ambient temperature, increases susceptibility to cold stress. ¹⁰ Cold exposure stimulates cutaneous cold receptors and increases catecholamine levels, which in turn trigger vasoconstriction and elevated blood pressure. ¹⁰ , ¹⁶ A hospital‐based study in Wuhan found that nighttime cold spells were more strongly associated with severe symptoms of coronary heart disease than daytime cold spells. ¹⁷ However, few studies have examined whether environmental factors such as air pollution or relative humidity modify the associations between nighttime cold exposure and cardiovascular outcomes. ¹⁸

This study therefore investigated the short‐term associations between cold exposure, separately for daytime and nighttime periods, and acute cardiovascular hospitalizations in Hong Kong, a subtropical city characterized by dry winters with a mean temperature of around 16°C. ¹⁹ Because of the mild climate, most buildings in Hong Kong are not equipped with central heating systems, and space heating accounts for only about 4% of household energy use during wintertime. ²⁰ We focused on 3 major cardiovascular outcomes: AMI, CHF, and cerebrovascular diseases. Attention was given to nighttime cold exposure and potential sex‐specific and age‐specific differences in vulnerability.

METHODS

Study Design and Setting

This study was conducted and reported in accordance with the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) guidelines for observational studies. ²¹ This was a territory‐wide ecological time‐series study in Hong Kong. We linked aggregated daily counts of emergency hospitalizations with citywide environmental measurements to examine short‐term associations between cold exposure and cardiovascular emergency hospitalizations from 2012 to 2022. The study protocol was approved by the Joint Chinese University of Hong Kong—New Territories East Cluster Clinical Research Ethics Committee (CRE‐2020.112). Since this study used aggregated data and did not directly involve individual participants, informed consent was not required. The data that support the findings of this study are available from the corresponding author upon reasonable request.

Data Sources and Outcome Definitions

Hospital admission data were obtained from the Clinical Data Analysis and Reporting System (CDARS) of the Hospital Authority, Hong Kong. CDARS is an electronic health care database that covers all public hospitals and clinics across a total of 18 districts in Hong Kong. All data in CDARS are anonymized to ensure confidentiality. The Hospital Authority is the primary public hospital service provider in Hong Kong and accounts for over 90% of all health care services provided to the local population. CDARS functions as an administrative database covering all admissions in public hospitals, ensuring comprehensive and continuous daily hospitalization records.

Daily emergency hospitalizations across all ages with a primary discharge diagnosis of AMI, CHF, and cerebrovascular diseases were retrieved from 2012 to 2022. Daily counts were aggregated from all public hospitals to generate territory‐wide totals for each disease category. Only unplanned admissions via accident and emergency departments were included. Cases were selected based on the principal discharge diagnosis, classified using the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD‐9‐CM): AMI (410), CHF (428), and cerebrovascular disease (430–438). Hospitalizations were analyzed in 3 groups by sex: total, men, and women, using data from all ages. For the age‐specific analyses, we focused on individuals aged 60 years and older (60–74, 75–89, and ≥90 years) because daily hospitalization counts among younger populations were too low to provide stable and reliable model estimates. The detailed subtypes of each disease can be found the in Table S1, Supporting Information.

Environmental Data

Meteorological data, including hourly ambient temperature (°C) and daily relative humidity (RH, %), were obtained from the Hong Kong Observatory, based on measurements from 53 weather stations from 2012 to 2022. Daytime was defined as 08:00–19:59 of the current day and nighttime as 20:00 of the previous day to 07:59 of the current day, consistent with local diurnal temperature patterns and human activity cycles, following prior epidemiological research that adopted a fixed 12‐hour division to reflect typical exposure behavior. ²² Hourly temperature was averaged across all stations to derive daytime and nighttime cold exposure metrics. Daily mean temperature was then calculated from these hourly values from 00:00 to 23:00 to align with the temporal resolution of emergency hospitalizations. Daily concentration of fine suspended particulates with a diameter ≤2.5 μm (PM_2.5) was collected from 15 general air quality monitoring stations operated by the Hong Kong Environmental Protection Department from 2012 to 2022. Daily mean PM_2.5 concentration was averaged across 14 general stations, excluding the Tap Mun station because of its remote mountaintop location, which may not accurately reflect population‐level exposure. Detailed station networks, including both weather and air quality monitoring stations, have been described in the previous study. ²³ Missing data for meteorological and air pollution variables at individual stations were first processed using an exponentially weighted moving average interpolation before averaging across stations.

Exposure Assessment: Daytime and Nighttime Cold Exposure Metrics

The Hong Kong Observatory defines a cold day as a day with a daily minimum temperature (the lowest recorded hourly temperature within a 24‐hour period) of 12°C or below, with an annual average of 15.2 days from 1991 to 2020. However, it does not define cold nights, despite having clear criteria for hot days and hot nights. Following previous studies, we adopted a widely used approach for quantifying daytime and nighttime thermal intensity. ²⁴ , ²⁵ The 5th percentile of daytime maximum temperature (T _max, 5th=15.96°C) and nighttime minimum temperature (T _min, 5th=11.23°C), based on data from 2012to 2022, were used as thresholds for identifying hourly cold exposure during the daytime (8:00–19:59) and nighttime (20:00–07:59) periods, respectively. To estimate the intensity of diurnal and nocturnal cold stress, we calculated the cold day excess (CDE, in °C · h) and cold night excess (CNE, in °C · h) as the cumulative temperature deviation of hourly records below their respective thresholds during the daytime and nighttime periods (Figure S1). For each daytime or nighttime period, all hourly temperature deviations below the threshold were summed to represent the cumulative cold intensity, while periods without any hourly records below the threshold were assigned a value of zero. A day (or night) was classified as a cold day (or cold night) when its CDE (or CNE) was greater than zero, indicating that at least 1 hourly temperature record during the daytime (or nighttime) period fell below the corresponding threshold. For example, exposure to 8°C for 1 hour during the night yields a CNE of 3.23°C · h. These daily indices (CDE and CNE) thus represented the total hourly‐integrated cold exposure during the daytime and nighttime periods of each daily cycle. Since these thresholds inherently captured the lower tail of the temperature distribution, most identified cold events occurred during the winter months. The equations used are as follows:

$${\mathrm{CDE}}_j=\sum _{i=1}^{12}\left(T_{\max ,5\mathrm{th}}-T_{\mathit{ij}}\right)\times I_{\mathrm{CDE}},I_{\mathrm{CDE}}=\left\{\begin{array}{c}0,\mathit{\text{if}}{T}_{\mathit{ij}}>T_{\max ,5\mathrm{th}}\\ 1,\mathit{\text{if}}{T}_{\mathit{ij}}\le T_{\max ,5\mathrm{th}}\end{array}\right.$$ (1)

(2)

where $T_{\mathit{ij}}$ is the hourly temperature at the $i^{\mathit{th}}$ hour of day $j$. The thresholds $T_{\max ,5\mathrm{th}}$ and $T_{\min ,5\mathrm{th}}$ represent the 5th percentiles of daily maximum and minimum temperature between 2012 and 2022, respectively.

Statistical Analysis

To reduce potential bias and temporal confounding, we used territory‐wide hospitalization records and adjusted all models for RH, PM_2.5, seasonality and long‐term trends. We applied a standard time‐series quasi‐Poisson regression model incorporating a distributed‐lag nonlinear model (DLNM) framework to estimate the exposure‐response associations between daily emergency hospitalizations for CVDs (ie, AMI, CHF, and cerebrovascular diseases) and 2 different dimensions of cold exposure (ie, CDE and CNE) in Hong Kong. ²² , ²⁶ Exposures were assigned at the city‐wide level by linking aggregated daily hospitalization counts with average environmental conditions across all monitoring stations. This approach follows previous temperature–health studies in Hong Kong and other city‐level analyses, ²² , ²⁷ where the compact geography and dense urban form lead to minimal spatial variability in temperature. The city‐wide time‐series design captures day‐to‐day variation and is well suited for assessing short‐term temperature–health associations in a densely built urban environment. The analysis covered 4018 daily observations from 2012 to 2022, during which daily hospitalization counts (summed across all public hospitals) were matched with corresponding daily meteorological and air pollution data averaged across monitoring stations. The optimum temperature for each CVD outcome was defined as the daily mean temperature corresponding to the lowest hospitalization risk, identified as the minimum point of the overall cumulative exposure‐response curve. Age‐specific analyses were conducted among individuals aged 60 to 74, 75 to 89, and ≥90 years, consistent with the stratification described in the data collection section, where hospitalization counts were sufficient to provide stable estimates. ²⁸

Separate models were constructed to estimate the general association between daily mean temperature and CVD hospitalizations (Equation 3) as well as the associations between specific cold exposure metrics (ie, CDE and CNE) and CVD risk (Equation 4). Each model used a similar quasi‐Poisson time‐series structure with a DLNM framework, with the exposure variable entered via a cross‐basis function. The exposure–response function for daily mean temperature was modeled nonlinearly using natural cubic splines, whereas for the cold‐exposure models (CDE and CNE), the exposure–response function was modeled linearly given their continuous nature as cumulative cold intensity measures. The lag–response dimension in the cold‐exposure models was modeled nonlinearly using a natural cubic spline with 4 degrees of freedom (df) to capture the delayed physiological association pattern related to cold exposure. Accordingly, for the cold‐exposure models, the reported cumulative relative risk (RR) represents the overall change in hospitalization risk associated with a fixed change in CDE/CNE over the full lag period. Based on prior evidence from Hong Kong, ²⁹ where heat effects were acute and disappeared within 1 week, while cold effects began on lag day 2 and persisted for 2 to 3 weeks, different maximum lag periods were applied. Specifically, a lag of 7 days was used for the daily mean temperature model to capture short‐term thermal impacts, and a lag of 21 days was applied to the cold‐exposure models to reflect the delayed and prolonged influence of cold. All models were adjusted for environmental covariates and included day of week as a categorical variable to control for weekly patterns in hospitalizations. In the general temperature–CVD model (Equation 3), the primary exposure was daily mean temperature. This model was adjusted for key environmental confounders, including same‐day RH and PM_2.5 concentration, which capture the dominant short‐term confounding contribution of these environmental factors. Prior time‐series studies have shown that RH can influence perceived temperature and affect health outcomes, ³⁰ while PM_2.5 is a recognized risk factor for health outcomes and is correlated with temperature. ³¹ Both variables are, therefore, commonly treated as confounders in temperature epidemiology studies. Natural cubic splines with 3 $\mathit{df}$ were used for RH and PM_2.5 to flexibly control their nonlinear effects, helping to isolate the association between temperature and CVD risk. Additionally, to control for seasonality and long‐term trends, a natural cubic spline of time with 8 $\mathit{df}$ per year was included in the model. ³² In the cold exposure‐specific models (Equation 4), each cold exposure metric was modeled separately using the same model formulation. RH and PM_2.5 were both incorporated as same‐day (lag 0) natural cubic spline terms (3 $\mathit{df}$) to account for their short‐term confounding influence while maintaining model stability. ²⁶ Cumulative RRs and their 95% CIs were derived from the fitted DLNM models by comparing the predicted risks at different exposure levels with the optimum temperature (for the temperature–CVD model, Equation 3) or with zero exposure (for the cold excess metrics, CDE/CNE, in Equation 4). For the cold‐exposure models, zero exposure indicates that no hourly temperature during the corresponding daytime or nighttime period fell below the defined cold threshold (ie, CDE/CNE=0). The reported RRs for Equation 4 were expressed per 10°C · h increment in cumulative cold exposure. The model equations are as follows:

$$\begin{array}{c}\log (\mu )=\alpha +\mathrm{cb}\left(T_{\text{mean}},\mathrm{lag}=7\right)+\mathrm{ns}\left(\mathrm{RH},\mathrm{df}=3\right)\hfill \\ +\mathrm{ns}\left({\mathrm{PM}}_{2.5},\mathrm{df}=3\right)+\mathrm{ns}\left(\text{Time},\mathrm{df}=8 per year\right)+\text{βDOW}\hfill \end{array}$$ (3)

$$\begin{array}{c}\log \left(\mu \right)=\alpha +\mathrm{cb}\left(\mathrm{CE},\mathrm{lag}=21\right)+\mathrm{ns}\left(\mathrm{RH},\mathrm{df}=3\right)\hfill \\ +\mathrm{ns}\left({\mathrm{PM}}_{2.5},\mathrm{df}=3\right)+\mathrm{ns}\left(\text{Time},\mathrm{df}=8 per year\right)+\text{βDOW}\hfill \end{array}$$ (4)

where $\mu =E\left(Y\right)$ represents the expected number of daily emergency hospitalizations for each CVD and $\alpha$ is the model intercept. $\mathit{\text{Time}}$ denotes continuous calendar time and was modeled using a natural spline with 8df per year to capture seasonality and long‐term trends, whereas $\mathrm{DOW}$ is a categorical variable for the day of the week, accounting for recurring weekly patterns across years. $T_{\text{mean}}$ is the daily mean temperature, whereas $\mathrm{CE}$ refers to CDE or CNE, modeled separately using the same model formulation. $\mathit{cb}\left(\right)$ is the cross‐basis function used to estimate the exposure–lag–response association, with the lag‐response dimension modeled using a natural cubic spline within the DLNM framework. For AMI and CHF, the exposure–response function for the daily mean temperature model was modeled using a natural cubic spline with 2 internal knots at the 10th and 90th percentiles of the daily mean temperature distribution. For cerebrovascular disease, 3 internal knots at the 10th, 50th, and 90th percentiles were used. The selection of knot placement and $\mathrm{df}$ was based on the minimizing the quasi‐Akaike information criterion and residual partial autocorrelation. ³³

We quantified the total number of emergency hospitalizations for CVDs associated with daytime and nighttime cold exposure (CDE and CNE, respectively) in Hong Kong from 2012 to 2022, based on daily aggregated counts within a citywide time‐series ecological design. The total attributable number and attributable fraction were estimated separately for CDE and CNE using the attrdl() function in R within the DLNM framework, as described by Gasparrini and Leone (2014). ³⁴ To account for uncertainty in the exposure–response association, we performed Monte Carlo simulations by drawing 5000 samples from a multivariate normal distribution of the model coefficients based on the estimated covariance matrix. ³⁴ This simulation approach assumes that the model coefficients, such as β, are jointly normally distributed, capturing their correlations as reflected in the covariance matrix while enabling us to generate 5000 sets of coefficients that reflect the variability in the estimated risks associated with CDE and CNE exposures for different CVD types and years in Hong Kong. For each simulation, the attributable fraction was calculated and then multiplied by the total number of hospitalizations to derive the corresponding attributable number. The empirical 95% CIs were obtained from the 2.5th and 97.5th percentiles of the simulated attributable numbers.

$${\mathrm{AF}}_i=1-\exp \left(-{\beta }_i\right)$$ (5)

$${\text{AN}}_i=N_i·{\mathrm{AF}}_i$$ (6)

where $i$ represents different CVD types. The parameter ${\beta }_i$ represents the log relative risk associated with CDE or CNE, estimated using the DLNM framework; $N$ represents the total number of hospitalizations for each CVD from 2012 to 2022.

Sensitivity Analyses

To evaluate the robustness of our results, we conducted sensitivity analyses with reference to previous studies. ³⁵ , ³⁶ First, we tested alternative lag structures for both temperature and cold‐exposure models to confirm that estimated risks remained stable as the lag window extended. Second, we also evaluated the sensitivity of results to different cold‐night and cold‐day definitions by applying alternative percentile thresholds including 1st percentile [the cold‐day threshold was T _max=11.95°C and the cold‐night threshold T _min=7.95°C] and 10th percentile [T _max=17.79°C and T _min=13.17°C] (Table S2). Third, we assessed model robustness for the CNE model by varying the df for the time variable (4, 6, and 10 $\mathit{df}$ per year, compared with 8 $\mathit{df}$ in the main model) and by sequentially removing environmental covariates, including PM_2.5 and RH. We further tested short‐lag specifications (0–3‐day moving averages) for RH and PM_2.5 to examine the potential influence of short‐term delayed confounding effects. In addition, we examined potential modification by PM_2.5 levels by stratifying analyses into high‐ and low‐pollution conditions. High‐pollution days were defined as those with daily mean PM_2.5 concentration ≥the 75th percentile (29.5 μg/m³), whereas low‐pollution days were defined as those ≤25th percentile (12.5 μg/m³).

Statistical significance of the differences between cold day and cold night associations was evaluated using the Wald test. ³⁷ All analyses and data preprocessing were conducted in R (version 4.4.1).

RESULTS

Descriptive Analysis

A total of 77 392 emergency hospitalizations for AMI, 159 782 for CHF, and 179 870 for cerebrovascular disease were recorded between 2012 and 2022. On average, women accounted for 47% of all emergency hospitalizations, while men accounted for 53%. The daily mean concentration of PM_2.5 was 22.8 μg/m³, with a maximum value of 115.6 μg/m³. The average daily mean temperature was 23.1°C, ranging from 3.7 to 32.2°C. Mean RH during the study period was 79.8%, with an SD of 10.7% (Table 1).

Table 1.

Descriptive Statistics of Daily Emergency Hospitalizations (Count) for AMI, Cerebrovascular Disease, and CHF, Temperature (°C), RH (%), and PM_2.5 (μg/m³) in Hong Kong During 2012 to 2022

Variable	Sex/Age	Mean (SD)	Min	1st	5th	25th	50th	75th	95th	99th	Max
AMI	Both	19 (6)	4	8	11	16	19	23	29	34	45
	Men	12 (4)	1	4	6	9	12	15	19	23	29
	Women	7 (3)	0	1	3	5	7	9	13	15	22
	Age 60–74 y	5 (2)	0	1	2	3	5	7	9	12	15
	Age 75–89 y	6 (3)	0	1	2	4	6	7	10	13	18
	Age ≥90 y	7 (4)	0	0	2	4	6	9	14	17	24
Cerebrovascular disease	Both	45 (8)	18	28	33	40	45	50	58	64	75
	Men	25 (5)	8	13	17	21	25	28	34	39	46
	Women	20 (5)	6	10	12	17	20	23	28	32	38
	Age 60–74 y	13 (4)	1	4	6	9	12	15	20	23	35
	Age 75–89 y	15 (4)	4	7	9	12	15	18	22	25	32
	Age ≥90 y	13 (6)	0	3	5	9	12	17	22	26	36
CHF	Both	40 (12)	11	19	23	31	38	47	63	75	102
	Men	18 (7)	3	6	9	13	17	22	31	37	46
	Women	21 (7)	5	9	12	17	21	26	35	42	56
	Age 60–74 y	5 (3)	0	0	1	3	5	6	10	12	21
	Age 75–89 y	12 (5)	0	3	5	9	12	16	22	27	35
	Age ≥90 y	21 (9)	2	6	9	15	20	26	38	48	66
Temperature	…	23.1 (5.2)	3.7	10.6	14	19	24.1	27.6	29.4	30.1	32.2
RH	…	79.8 (10.7)	23.7	43	59.2	75.3	81.7	86.9	93.3	95.7	98.2
PM2.5	…	22.8 (14.3)	2.7	5	6.6	12.5	19.5	29.5	50	70.7	115.6

Values are presented as mean (SD), minimum, maximum, and selected percentiles.

AMI indicates acute myocardial infarction; and CHF, congestive heart failure.

The long‐term trends of emergency hospitalizations and environmental variables showed that daily emergency hospitalizations for all 3 CVDs remained relatively stable from 2012 to 2019, followed by a notable decline during the COVID‐19 period (2020–2022) (Figures S2a–c). Temperature showed a clear seasonal cycle with a slight but statistically significant increasing trend over time (P<0.001) in the context of global warming (Figure S2d). RH remained stable with no significant trend (Figure S2e). In contrast, PM_2.5 concentration showed a significant declining trend throughout the decade (slope=−0.0047, P<0.001) (Figure S2f). These temporal patterns provide important context for evaluating the health effects of environmental exposures in this study.

To investigate differences in environmental and health characteristics between daytime cold exposure and nighttime cold exposure, we compared distributions of cardiovascular emergency hospitalizations and key environmental variables (Figure 1). Cold days and nights were defined using the 5th percentile thresholds for daytime maximum temperature (15.96°C) and nighttime minimum temperature (11.23°C), as described in the Methods and applied consistently throughout the analyses. A total of 714 cold days and 201 cold nights were identified during the study period. The median cold intensity was higher for cold nights (7.1°C · h) than for cold days (6.4°C · h), suggesting higher cold intensity at night (Table S2). Most cold events occurred during the winter months (December–February). Cold nights were more concentrated and intense, extending slightly into March, whereas cold days occasionally appeared in November, April, and even May (Figure S3). Cold nights were associated with significantly lower temperature and humidity (Figure 1D through 1E). Dry air has been associated with increased mortality risks during cold weather. ³⁸ Emergency hospitalizations for AMI and CHF were also significantly higher during cold nights, while cerebrovascular hospitalizations showed no significant difference (Figure 1A through 1C). PM_2.5 concentration showed no significant difference between cold days and nights (Figure 1F). However, PM_2.5 concentration was typically elevated during winter in Hong Kong because of transboundary pollution and stable weather conditions. ²³ , ³⁹ Since both cold days and nights tended to occur in winter, the consistently high PM_2.5 levels throughout wintertime may contribute to the cardiovascular risk. These findings highlight the importance of assessing nighttime cold exposure in environmental health research while appropriately controlling for potential confounding effects by air pollution and humidity.

Figure 1. Comparison of emergency hospitalizations for AMI (A), cerebrovascular disease (B), and CHF (C), as well as daily mean temperature (D), relative humidity (E), and PM_2.5 concentration (F) between cold days and cold nights in Hong Kong during 2012 to 2022.

The statistical significance of the difference between cold days and cold nights was calculated using 2‐sided t‐tests. AMI indicates acute myocardial infarction; and CHF, congestive heart failure.

Exposure‐Response Associations of CVD With Daily Mean Temperature

To establish a baseline understanding of how temperature is associated with cardiovascular health, we first examined the exposure–response associations between daily mean temperature and emergency hospitalizations for AMI, cerebrovascular disease, and CHF. This analysis helps characterize the overall temperature‐related risk patterns and the optimum temperature for each cardiovascular emergency hospitalization outcome, setting the stage for a more focused investigation of cold exposure. In Figure 2, all 3 cardiovascular outcomes showed clear associations with cold temperature, with hospitalization risks increasing as temperature declines. The cold‐related risk was most pronounced for CHF, moderate for AMI, and weakest for cerebrovascular disease, whereas the heat‐related risks were relatively marginal across all 3 outcomes. The optimum temperature, defined as the temperature associated with the lowest hospitalization risk, was approximately 27.2°C for AMI, 28.7°C for cerebrovascular disease, and 31.7°C for CHF. Compared with their respective optimum temperatures, the RR during extreme cold (1st percentile, 10.6°C) was 1.98 (95% CI: 1.83–2.14) for CHF, 1.48 (1.41–1.55) for AMI, and 1.13 (1.09–1.18) for cerebrovascular disease. In contrast, extreme heat (99th percentile, 30.1°C) was associated with much smaller and statistically nonsignificant increases in risk, with RRs of 1.01 (0.98–1.05) for CHF, 1.03 (0.99–1.06) for AMI, and 1.01 (0.98–1.03) for cerebrovascular disease (Table 2). This cold‐dominant pattern aligned with previous studies focusing on cardiovascular outcomes in Hong Kong, where a marginal risk was observed at high temperature. ²⁹ , ⁴⁰ These results suggest that cold exposure is strongly associated with acute cardiovascular hospitalizations and highlight the need to further investigate the potential role of nighttime cold conditions.

Figure 2. Estimated exposure–response relationships between daily mean temperature and emergency hospitalizations for AMI (A), cerebrovascular disease (B), and CHF (C).

Curves represent cumulative RRs over lag 0 to 7 days, adjusted for long‐term trends, seasonality, relative humidity, and PM_2.5. Shaded areas indicate 95% CIs. Histograms at the bottom of each panel show the distribution of daily mean temperature in Hong Kong from 2012 to 2022. AMI indicates acute myocardial infarction; and CHF, congestive heart failure.

Table 2.

RRs of Emergency Hospitalizations for AMI, Cerebrovascular Disease, and CHF Associated With Daily Mean Temperature at a 7‐Day Lag in Hong Kong, 2012 to 2022

Disease	Temperature category	Relative risk
AMI (OP=27.2°C)	Extreme cold	1.48 (1.41–1.55)
	Moderate cold	1.28 (1.25–1.31)
	Moderate heat	1.01 (0.99–1.03)
	Extreme heat	1.03 (0.99–1.06)
Cerebrovascular disease (OP=28.7°C)	Extreme cold	1.13 (1.09–1.18)
	Moderate cold	1.08 (1.06–1.09)
	Moderate heat	1 (0.996–1.001)
	Extreme heat	1.01 (0.98–1.03)
CHF (OP=32.2°C)	Extreme cold	1.98 (1.83–2.14)
	Moderate cold	1.73 (1.64–1.83)
	Moderate heat	1.03 (0.99–1.08)
	Extreme heat	1.01 (0.98–1.05)

RRs are calculated for different temperature percentiles compared with the optimal temperature (OP). Values in parentheses indicate 95% CIs. Extreme cold: the 1st percentile of temperature (10.6°C). Moderate cold: the 10th percentile of temperature (15.8°C). Moderate heat: the 90th percentile of temperature (29.0°C). Extreme heat: the 99th percentile of temperature (30.1°C). AMI indicates acute myocardial infarction; and CHF, congestive heart failure.

Exposure‐Response Associations of CVD With CDE and CNE

Given the significant associations observed in the daily mean temperature analysis, we further investigated the specific associations between cold exposure and cardiovascular emergency hospitalizations under daytime and nighttime conditions. Figure 3 presents the associations of CDE and CNE with emergency hospitalizations for CVDs. Overall, CNE showed significantly stronger associations than CDE across all CVD emergency outcomes. However, this pattern was less consistent for cerebrovascular disease, where the difference between CDE and CNE was not statistically significant, particularly among men or among individuals younger than 90 years old. Notably, CHF exhibited the strongest association with nighttime cold, whereas cerebrovascular disease showed the weakest association. The RR for CHF emergency hospitalizations was 1.44 (1.40–1.48) for CNE, compared with 1.19 (1.18–1.21) for CDE (Figure 3C). By contrast, the RR for cerebrovascular disease was 1.06 (1.04–1.08) for CNE, compared with 1.03 (1.02–1.03) for CDE (Figure 3B). Sex‐stratified analysis of nighttime cold exposure showed no statistically significant sex differences for AMI (P=0.485) or cerebrovascular disease (P=0.353), whereas the association with CNE was significantly stronger in men than in women for CHF at the 0.05 level (P=0.037), as shown in Table S3. For instance, among CHF emergency cases (Figure 3C), the RR for men was 1.48 (1.43–1.53), while for women it was 1.40 (1.36–1.45) during nighttime cold exposure. By contrast, age‐stratified analysis showed that the ≥90 age group had significantly higher RRs than the 60 to 74 age group across all 3 cardiovascular outcomes under nighttime cold exposure. For example, under CNE, the RR for CHF was 1.51 (1.47–1.56) in the population ≥90 years old, compared with 1.23 (1.16–1.31) in those aged 60 to 74 years old (P<0.001). These findings highlight the statistical significance and varying clinical relevance of associations between cold exposure and cardiovascular emergency outcomes, while also indicating the stronger associations for nocturnal cold and the importance of considering vulnerability by sex and age group in cold weather health risk assessments.

Figure 3. Forest plot of cumulative risk ratios and 95% CIs for emergency hospitalizations of AMI (A), cerebrovascular disease (B), and CHF (C) associated with cold day excess (CDE, blue diamonds) and cold night excess (CNE, red diamonds), stratified by sex (both sexes, men, women) and age group (60–74, 75–89, ≥90 years).

Statistical significance for the difference between CDE and CNE was calculated based on the Wald test. RRs represent the change in hospitalization risk associated with each 10°C · h increment in cumulative cold exposure relative to zero cold exposure (ie, CDE/CNE=0) for CDE and CNE, respectively. AMI indicates acute myocardial infarction; and CHF, congestive heart failure.

Health Burden for CVDs Associated With CDE and CNE

Figure 4 presents the estimated total number of emergency hospitalizations for CVDs associated with daytime and nighttime cold exposure in Hong Kong from 2012 to 2022. Overall, the health burden associated with CDE was higher than that associated with CNE, reflecting the larger number of cold days (714 days) compared with cold nights (201 nights). The total number of emergency hospitalizations estimated to be associated with CDE across all 3 CVDs was approximately 13 575 (95% CI: 12442–14 654), whereas 6805 (6045–7536) were estimated to be associated with CNE (Table 3). Despite fewer cold nights, the estimated burden associated with nighttime cold remained substantial, consistent with the stronger RRs observed for CNE than for CDE. Among all CVDs, CHF showed the highest estimated burden associated with CNE, with an estimated 4727 (4406–5035) emergency cases in the total population during nighttime cold exposure (Figure 4C). This was followed by AMI with 1355 (1183–1517) emergency cases and cerebrovascular disease with 723 (456–984) emergency cases. The patterns of sex differences varied by disease type. For AMI, men showed a substantially higher number of hospitalizations associated with CNE than women, with 879 (749–1002) emergency cases compared with 472 (372–569) emergency cases, respectively (Figure 4A). In contrast, sex differences were marginal for both CHF and cerebrovascular disease (Figure 4B through 4C). For CHF, the estimated burden was 2347 (2164–2519) emergency cases in men and 2375 (2179–2570) in women, which may reflect the higher baseline number of CHF admissions among women despite the stronger risk observed in males (Table 1). For cerebrovascular disease, men and women each accounted for approximately 334 (142–528) and 387 (215–554) emergency cases, respectively. The estimated burden associated with nighttime cold was higher in older age groups, with the ≥90 age group showing the highest estimated burden. Specifically, for the ≥90 age group, CHF accounted for 3112 (2903–3315) emergency cases, AMI for 752 (656–843) emergency cases, and cerebrovascular disease for 598 (464–733) emergency cases (Figure 4D through 4F). These findings highlight the substantial estimated health burden associated with nighttime cold exposure across CVD types, while also emphasizing the importance of considering disease‐specific, sex‐specific, and age‐specific vulnerabilities in public health planning and response.

Figure 4. Estimated total number of emergency hospitalizations associated with nighttime cold exposure for AMI (A, D), cerebrovascular disease (B, E), and CHF (C, F), stratified by sex in (A–C) (both sexes, women, and men), and by age group in (D–F) (60–74, 75–89, and ≥90 years) during 2012 to 2022 in Hong Kong.

Error bars represent 95% empirical CIs derived from Monte Carlo simulations (n=5000). AMI indicates acute myocardial infarction; and CHF, congestive heart failure.

Table 3.

Estimated Attributable Fraction (AF, %) and Attributable Number (AN) of Emergency Hospitalizations Associated With Daytime and Nighttime Cold Exposure for AMI, Cerebrovascular Disease, and CHF, Stratified by Sex (Both Sexes, Women, and Men) and Age Group (60–74, 75–89, and ≥90) During 2012 to 2022 in Hong Kong

Disease	Category	Cold day AF	Cold day AN	Cold night AF	Cold night AN
AMI	Both	3.51% (3.19%–3.83%)	2719 (2467–2966)	1.75% (1.53%–1.96%)	1355 (1183–1517)
	Women	3.44% (2.91%–3.93%)	980 (830–1122)	1.66% (1.31%–2%)	472 (372–569)
	Men	3.56% (3.16%–3.94%)	1739 (1547–1925)	1.8% (1.53%–2.05%)	879 (749–1002)
	Age: 60–74 y	1.59% (0.98%–2.19%)	324 (199–447)	0.78% (0.37%–1.18%)	159 (76–240)
	Age: 75–89 y	3.97% (3.4%–4.53%)	912 (783–1042)	1.95% (1.57%–2.32%)	448 (362–533)
	Age: ≥90 y	5.53% (4.98%–6.04%)	1473 (1327–1609)	2.82% (2.46%–3.16%)	752 (656–843)
Cerebrovascular disease	Both	0.76% (0.52%–0.97%)	1362 (942–1747)	0.4% (0.25%–0.55%)	723 (456–984)
	Women	0.96% (0.63%–1.28%)	767 (501–1026)	0.48% (0.27%–0.69%)	387 (215–554)
	Men	0.59% (0.29%–0.89%)	586 (287–891)	0.33% (0.14%–0.53%)	334 (142–528)
	Age: 60–74 y	0% (0%–0.004%)	0 (0–2)	0% (0%–0.13%)	0 (0–70)
	Age: 75–89 y	0.74% (0.36%–1.12%)	448 (216–678)	0.4% (0.17%–0.64%)	246 (104–386)
	Age: ≥90 y	2.26% (1.84%–2.66%)	1162 (946–1371)	1.16% (0.9%–1.42%)	598 (464–733)
CHF	Both	5.94% (5.65%–6.22%)	9494 (9033–9941)	2.96% (2.76%–3.15%)	4727 (4406–5035)
	Women	5.57% (5.24%–5.91%)	4847 (4559–5138)	2.73% (2.5%–2.95%)	2375 (2179–2570)
	Men	6.38% (6.02%–6.73%)	4645 (4383–4897)	3.22% (2.97%–3.46%)	2347 (2164–2519)
	Age: 60–74 y	3.26% (2.62%–3.87%)	635 (509–752)	1.46% (1.03%–1.87%)	285 (201–363)
	Age: 75–89 y	5.36% (4.96%–5.76%)	2662 (2462–2859)	2.64% (2.35%–2.91%)	1310 (1166–1445)
	Age: ≥90 y	7.16% (6.82%–7.49%)	6146 (5851–6429)	3.63% (3.38%–3.86%)	3112 (2903–3315)

Values in parentheses indicate 95% CIs derived from Monte Carlo simulations (n=5000). AMI indicates acute myocardial infarction; and CHF, congestive heart failure.

Sensitivity Test

We examined the lag–response associations to determine appropriate lag windows for the temperature and cold‐exposure models. For the daily mean temperature model, the cumulative RR appeared to peak within the first few days and plateau by approximately 7 days for AMI and CHF, while the RR for cerebrovascular disease flattened earlier, around 3 days (Figure S4). These findings support the use of a short lag window for the general temperature–CVD model. In contrast, for the CNE model, the RR accumulated over a longer period (Figure S5). The RR for AMI and cerebrovascular disease increased steadily and tended to plateau around 3 weeks, with cerebrovascular disease showing a slight decline thereafter. For CHF, the RR associated with cold exposure continued to rise across the 30‐day lag period without clear attenuation. Based on these lag‐response patterns, we selected a 7‐day lag for the daily mean temperature model and a 21‐day lag for the cold‐exposure models to adequately capture both short‐term and delayed cold‐related associations.

Sensitivity analyses using alternative percentile thresholds for defining cold nights and cold days (1st and 10th percentiles) showed that the estimated RRs increased consistently with increasing cold intensity (Table S4). When a more extreme definition was applied (1st percentile, 7.95°C for cold nights), the RRs were markedly higher, whereas the milder 10th percentile threshold (13.17°C) produced smaller yet consistent estimates. This graded pattern indicated a consistent exposure–response association between cold intensity and CVD risk. The 5th percentile threshold was retained because it aligned with widely used definitions in temperature–health research, ²² , ²⁴ , ²⁵ , ³⁵ reflecting locally adapted and epidemiologically meaningful cold conditions. This choice balances representativeness and statistical stability while ensuring comparability with prior multi‐city and regional studies.

The cumulative RRs of emergency hospitalizations for AMI, cerebrovascular disease, and CHF associated with CNE remained robust across a range of alternative model specifications. As shown in Table S3, the estimated RRs were largely unchanged when the temporal $\mathit{df}$ was varied from 4 to 10 per year. Although removing PM_2.5 or RH from the model altered the cumulative RRs marginally, the effect modification by PM_2.5 levels was statistically significant (P<0.001). When stratified by PM_2.5 levels, cold‐night RRs were substantially higher under high‐pollution conditions while protective effects were observed for all 3 CVDs under low‐pollution conditions. These findings implied that PM_2.5 could amplify cold‐related CVD risks under polluted conditions. This pattern aligned with the environmental context in Hong Kong, where winter cold surges often coincide with transboundary air pollution events from the Pearl River Delta. ³⁹ Additional sensitivity analyses that tested alternative lag specifications for environmental covariates showed that applying a 0–3‐day moving average for RH and PM_2.5 did not materially change the estimated associations. These results indicated that the observed associations were not sensitive to alternative lag specifications for RH and PM_2.5. Therefore, the lag‐0 specification for environmental covariates was retained in the main analysis to ensure model stability and interpretability.

DISCUSSION

To our knowledge, this is the first study to comprehensively assess the short‐term associations of nighttime cold exposure with acute cardiovascular hospitalizations in a typical subtropical city. This study indicated that nighttime cold exposure is an important factor associated with cardiovascular health in Hong Kong, with significantly higher RRs of emergency hospitalizations for AMI, CHF, and cerebrovascular disease compared with daytime cold exposure, particularly for CHF. These associations were especially elevated among population in the oldest age group. Despite cold nights occurring less frequently than cold days, our model‐based burden analysis suggested that 6805 (6045–7536) excess CVD emergency hospitalizations were associated with nocturnal cold weather during 201 nights during the study period. These findings emphasized the importance of incorporating nighttime thermal stress into public health surveillance and intervention strategies.

Our findings aligned with the growing body of global evidence demonstrating that cold temperatures are associated with increased cardiovascular morbidity and mortality compared with heat. A multi‐country study involving 384 locations across 13 countries reported that low temperature accounted for a significantly higher proportion of temperature‐related deaths (7.29%) compared with high temperature (0.42%). ⁴¹ Similar conclusions have been drawn in national‐level studies: in Spain, 11.62% of deaths were attributable to low temperature versus 2.71% for high temperature, ²⁸ while in China, total cold‐related deaths in 2019 exceeded 580 000, more than 40 times the number of heat‐related deaths, with CVD comprising the majority of these cases. ⁴² Consistent with this global pattern, our findings showed that cold exposure was significantly associated with increased emergency hospitalizations for all cardiovascular outcomes, particularly for CHF. Compared with their respective optimum temperatures, the RRs during extreme cold were 1.98 (1.83–2.14) for CHF, 1.48 (1.41–1.55) for AMI, and 1.13 (1.09–1.18) for cerebrovascular disease. Although high temperature has been reported to increase CVD admissions in other regions, the heat‐related risks observed in this study were relatively marginal compared with the risks associated with cold exposure. This pattern aligned with findings from other tropical and subtropical cities such as Hong Kong, Taiwan, and Vietnam, ¹⁴ , ²⁹ , ⁴⁰ , ⁴³ where summers are long, hot, and humid. The limited heat associations in these settings may reflect high population adaptation, the high prevalence of air conditioning, and reduced outdoor activity during hot weather. In Hong Kong, most residents spend extended periods in air‐conditioned indoor environments, which may reduce heat exposure. In contrast, cold periods are short, and most households lack heating systems, making cold weather a more physiologically stressful condition for the population. Previous research has shown declining heat‐related mortality and increasing heat resilience over time, particularly among older adults. ⁴⁴ , ⁴⁵ Moreover, heat associations are typically short‐lived (lag 0–1 days) and tend to attenuate over longer periods because of harvesting effects. ²⁹ , ⁴³

Our study emphasized that nocturnal cold was associated with adverse health outcomes. Epidemiological studies suggest that cold nights can induce sharper rises in adverse events compared with cold days. For example, a nationwide analysis in China found that a nighttime‐only cold spell increased all‐cause mortality by about 11.9%, compared with 7.8% for a daytime‐only cold spell. ³⁸ Similarly, a multi‐city study of mental health hospitalizations reported higher risk during cold nights than cold days, with attributable fractions of 25.7% and 19.7%, respectively. ⁴⁶ While these findings demonstrated the vulnerability of these populations to cold weather, which may be stronger during nighttime, studies focusing on subtropical cities remain limited. This may be because residents in these regions are generally more concerned about heat‐related threats, given the prevalence of high temperature throughout much of the year. For example, although Hong Kong has introduced various climate resilience initiatives, including research funding, policy development, and urban planning, existing strategies remain primarily focused on heat‐related risks, particularly hot days and hot nights, which have been well studied in prior studies. ²² , ⁴⁷ In contrast, the health risks associated with nighttime extreme cold spells have received far less attention. However, growing evidence suggests that cold exposure can induce hidden physiological stress at night by disrupting thermoregulatory and cardiovascular functions during sleep. ⁴⁸ , ⁴⁹ Our study addressed this gap by showing that nighttime cold exposure was associated with stronger hospitalization risks for cardiovascular conditions than daytime cold exposure in a typical subtropical city. During sleep, core body temperature naturally declines because of circadian rhythms and reduced metabolic heat production, reaching its lowest point in the early morning hours. ⁵⁰ The structure of sleep is sensitive to ambient temperature; and cold exposure at night may further disrupt thermal balance and affect physiological stability. ⁵¹ In response, the body activates stress‐related mechanisms: cold exposure increases sympathetic nervous activity and vasoconstriction of blood vessels, both of which elevate blood pressure and cardiac workload. ⁵² Additionally, cold exposure can stimulate brown adipose tissue thermogenesis through β3‐adrenergic signaling, driven by sympathetic nervous system activation. This response increases heart rate, vasoconstriction, and cardiac workload. In individuals with pre‐existing CVD conditions, such effects may increase myocardial oxygen demand, potentially triggering ischemic events or worsening existing cardiac conditions. ⁵² , ⁵³ , ⁵⁴ Nevertheless, more research is needed to elucidate the physiological mechanisms underlying cold‐induced cardiovascular events and to explore strategies for mitigating these risks, especially during nighttime in subtropical cities.

This study had several strengths that enhance its scientific and practical value. It is among the first to specifically examine the short‐term associations of nighttime cold exposure with emergency cardiovascular hospitalizations in a subtropical city, using a detailed classification of daytime and nighttime periods. By separately analyzing cold day and cold night exposures, the study showed substantially stronger associations for nighttime cold, which has rarely been addressed in previous research, particularly in subtropical settings. This distinction is important because most prior studies have focused on daily average temperature and overlooked the potential temporal differences in thermal stress. Another major strength lies in the use of high‐resolution daily emergency hospitalization data spanning over a decade, which allowed for the robust estimation of acute associations. Stratified analyses by disease type and sex further enriched the findings, and adjustment for key confounders such as PM_2.5 and RH ensured the reliability of the results. From a public health and climate adaptation perspective, these findings underscored the importance of recognizing cold‐related health risks even in subtropical regions. Although cold nights are relatively infrequent, the estimated burden associated with cold nights is substantial. During the study period, over 201 cold nights were identified, resulting in more than 6000 emergency cardiovascular hospitalizations associated with nighttime cold exposure. These results can support predictive modeling of surges in cardiovascular hospital admissions, helping hospital administrators and clinicians better prepare when cold nighttime weather is anticipated. As climate change intensifies, traditionally subtropical areas must not overlook cold‐related health risks. Integrating these findings into urban planning, elderly care strategies, and early warning systems can help build more climate‐resilient and inclusive cities. Future research should further identify spatial hotspots and quantify the long‐term health and economic burden of nighttime cold exposure, and project its future estimated burden under climate change scenarios to support targeted adaptation and long‐term policy planning.

Although the time‐series design effectively controlled for major time‐independent variables, such as calendar day, and day of week, this study still had several limitations that should be acknowledged. First, some other potential modifiers, such as influenza epidemics, which can transiently increase AMI admissions, ⁵⁵ and socioeconomic inequality, which may contribute to baseline vulnerability, ⁵⁶ were not available in this study. In addition, although measured PM_2.5 and RH were adjusted for in all models, residual confounding from air pollution exposure and other unmeasured time‐varying factors associated with both cold conditions and cardiovascular events cannot be entirely excluded. Nevertheless, because many of these factors tended to vary gradually over time, part of their variation may have been captured by the temporal adjustment functions in the time‐series model. Second, exposure to cold was based on ambient temperature from fixed‐site monitoring stations, which may not fully represent indoor conditions. However, Hong Kong does not rely on heating during its short winter season, and most homes lack heating systems, so indoor and outdoor temperatures are generally similar during cold spells. Hong Kong Housing Authority has confirmed that indoor temperature closely follows outdoor conditions. Patients experiencing acute CVD can also reasonably be assumed to have been exposed to ambient cold conditions, either directly outdoors or indirectly through unheated indoor environments. Third, while we used emergency hospitalization data from all public hospitals in Hong Kong, private hospital admissions were not included; however, according to the Hong Kong Domestic Health Accounts, public hospitals managed by the Hospital Authority provide about 90% of all inpatient services. Furthermore, the Hong Kong Health Bureau reports that the public sector dominates acute and emergency care, managing around 27 800 hospital beds compared with about 4200 in the private sector. The dual‐track health care system in Hong Kong provides heavily subsidized emergency and inpatient services through public hospitals, whereas private hospitals primarily focus on elective or specialty care. Therefore, any bias from excluding private hospital admissions is likely small and would lead to a conservative underestimation of the overall estimated health burden associated with cold exposure. Fourth, although we relied on principal diagnosis codes to identify cardiovascular events, which are generally more specific for acute admissions, potential misclassification remains a limitation; however, the use of the CDARS database, recognized for standardized and consistent coding across all public hospitals, helps ensure reliability. While sex‐specific differences were explored, other demographic and clinical subgroups, such as comorbidities and medication use, could not be analyzed because of data limitations and should be addressed in future work to refine risk stratification and public health response. Finally, as this was a citywide time‐series ecological study, the associations reflect population‐level relationships and may be subject to ecological bias or the ecological fallacy, though such bias is likely minimal given the consistent exposure across the population and the use of established analytical methods.

CONCLUSIONS

This study enabled a precise temporal evaluation of cold‐related cardiovascular associations by leveraging over a decade of high‐resolution daily emergency hospitalization data from all public hospitals in Hong Kong. After adjustment for measured environmental confounders such as air pollution and RH, our analysis provided a context‐specific assessment of the association between extreme nighttime cold and cardiovascular emergency hospitalizations. The findings are expected to enhance our understanding of population susceptibility to cold stress in subtropical settings, supporting the development of targeted public health interventions in response to climate change, such as improving nighttime indoor thermal comfort through better housing insulation and bedding, promoting public awareness of nighttime temperature drops, and implementing community‐based outreach or energy support programs for vulnerable populations, especially older adults. Given that most buildings in Hong Kong are not equipped with central heating systems, enhancing household preparedness and indoor thermal conditions may be especially critical to mitigate the cardiovascular burden of nighttime cold under a changing climate.

Sources of Funding

This work was supported by the Ministry of Education, Singapore, under the MOE Academic Research Fund (AcRF) Tier 3 Award MOET32022‐0006 and the MOE AcRF Tier 1 Project (RT2/22).

Disclosures

All the authors declare no conflicts of interest.

Supporting information

Figures S1–S5

Tables S1–S4

STROBE Statement